Reproduce “LAION-Audio-XL: A Web-Scale Corpus and Scaling Studies for Contrastive Language-Audio Pretraining” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Gijs Wijngaard, Felix Förster, Ross Wightman, Marianna Nezhurina, Christoph Schuhmann, Jenia Jitsev, A. Sophia Koepke
The authors introduce LAION-Audio-XL, a curated audio-text corpus, and train CLAP2 from scratch to study contrastive audio-language scaling. The final corpus contains about 366 million clips, and tests cover five model sizes and sample exposures up to 1,024 million. CLAP2-L scores 53.6% on MAEB-30; fitted compute allocation favors sample-count growth, but boundary-limited fits make the trend indicative, not definitive.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.